254 results about "Expectation–maximization algorithm" patented technology

A composite language model may include a composite word predictor. The composite word predictor may include a first language model and a second language model that are combined according to a directed Markov random field. The composite word predictor can predict a next word based upon a first set of contexts and a second set of contexts. The first language model may include a first word predictor that is dependent upon the first set of contexts. The second language model may include a second word predictor that is dependent upon the second set of contexts. Composite model parameters can be determined by multiple iterations of a convergent N-best list approximate Expectation-Maximization algorithm and a follow-up Expectation-Maximization algorithm applied in sequence, wherein the convergent N-best list approximate Expectation-Maximization algorithm and the follow-up Expectation-Maximization algorithm extracts the first set of contexts and the second set of contexts from a training corpus.

The invention provides a forest point cloud sorting method based on mode recognition, and belongs to the research field of forest canopy structure parameter obtaining methods. The method comprises the following steps of: obtaining three-dimensional laser point cloud data of the forest, selecting training sample of each sort, and computing obvious feature of each training sample; initially sorting the point cloud data by using an algorithm for obtaining Gauss mixed model based on expectation maximization algorithm, sorting the initial sorting result by using a filter and then processing. Compared with the traditional and existing method of obtaining the canopy structure parameter by using LIDAR, the method provided by the invention is time-saving and labor-saving, free from damaging the forest canopy structure, strong in applicability and high in precision.

A system and method of extracting collocation translations is presented. The methods include constructing a collocation translation model using monolingual source and target language corpora as well as bilingual corpus, if available. The collocation translation model employs an expectation maximization algorithm with respect to contextual words surrounding collocations. The collocation translation model can be used later to extract a collocation translation dictionary. Optional filters based on context redundancy and/or bi-directional translation constrain can be used to ensure that only highly reliable collocation translations are included in the dictionary. The constructed collocation translation model and the extracted collocation translation dictionary can be used later for further natural language processing, such as sentence translation.

A maximum likelihood (ML) linear regression (LR) solution to environment normalization is provided where the environment is modeled as a hidden (non-observable) variable. By application of an expectation maximization algorithm and extension of Baum-Welch forward and backward variables (Steps 23a–23d) a source normalization is achieved such that it is not necessary to label a database in terms of environment such as speaker identity, channel, microphone and noise type.

The present invention discloses an estimation method of a quasi-stationary broadband array signal direction of arrival (DOA) based on block sparse Bayesian learning. An intra-frame correlation and aninterframe independence of a quasi-stationary broadband signal frequency spectrum are employed to set a corresponding a block sparse prior distribution model for signals, and a block sparse Bayesian model is employed to perform estimation of sparse signals, so that an estimation result with higher precision is obtained. Array receiving signals are subjected to appropriate framing processing, eachframe of the signal is subjected to Fourier transform, and each block sparse Bayesian model for each signal is established in a frequency domain; under an assumption of each frame of the signal is independent, information of all the frames is combined to establish a total Bayesian model, and hyper-parameter vectors are employed to control a sparsity of all the frames of signals to be reconstructed; and finally, the expectation maximization algorithm (EM) is employed to obtain an iterative update formula of the hyper-parameter vectors. The estimation method of a quasi-stationary broadband arraysignal direction of arrival based on block sparse Bayesian learning fully utilizes a short-time stability feature of quasi-stationary broadband array signals to establish a block sparse model, and therefore a higher DOA estimation precision can be obtained.

The invention discloses a method for uniformly adjusting a multicolor picture based on color transmission. The method comprises the following steps: 1, one picture with two colors is divided into two pictures, namely a reference picture and a picture to be adjusted; and the two pictures are converted to a 1 alpha beta color space; 2, the two pictures are divided through a K mean algorithm; and each partition region of the reference picture is sampled; 3, an expectation maximization algorithm is used to carry out probability partition on the two pictures respectively; and a mapping relation is established between each partition region of the reference picture and each partition region of the picture to be adjusted; 4, the picture to be adjusted is subjected to brightness zooming in order to keep consistency in brightness of the reference picture; and 5, a point of sampling points of the reference picture which is most matched with each point of the picture to be adjusted is searched out; and the color value of the most matched point of the reference picture is transmitted to the corresponding point of the picture to be adjusted. The method has small calculation amount, is convenient and rapid and solves the problem of color discontinuity in the prior multicolor picture adjustment method.

The invention discloses a speaker recognition method based on a Gaussian mixture model (GMM) embedded with a time delay neural network (TDNN). In the speaker recognition method, the advantages of the TDNN and the GMM are fully considered, the TDNN is embedded into the GMM, and solves a residual of input and output vectors of the TDNN by fully utilizing the time sequence of an input characteristic vector through the conversion of a time delay network, and the residual modifies the training of the GMM through an expectation maximization method; besides, a likelihood probability is acquired by a modified GMM model parameter and the residual, and a TDNN parameter is modified by an inertial backward inversion method so as to ensure that parameters of the GMM and the TDNN are alternately updated. An experiment shows that: a recognition rate of the method is improved to a certain extent compared with that of a baseline GMM under various signal to noise ratios.

The invention provides a residual life prediction method for a non-stationary degradation process with uncertain impact and belongs to the technical field of reliability engineering. The impact effect is combined with a continuous degradation process, thereby more reasonably describing the actual degradation process. In a state estimation phase, a new online state estimation algorithm is provided to identify the real system state and to provide a necessary support for subsequent residual life prediction. In a parameter estimation stage, unknown parameters in a model are obtained by adopting an expectation maximization algorithm. For residual life production, the uncertainty of state estimation and the influence of the impact effect are considered at the same time, and an expression is analyzed to give a probability density function and a cumulative probability density function of residual life prediction distribution. The provided model meets the actual degradation condition, a more accurate residual life prediction result can be obtained, and the residual life prediction method is of great significance in fault prediction and health management in engineering.

The invention discloses a remaining life prediction method based on two-stage random degradation modeling, and belongs to the field of industrial monitoring and fault diagnosis. The method mainly comprises the step of off-line modeling and the step of on-line parameter updating and remaining life prediction; wherein the step of off-line modeling comprises the substeps that historical degradation data is collected; maximum likelihood estimation is used for obtaining a change-point estimation value of each group of degradation, and the distribution characteristics of change-points are obtained through statistic analysis; two-stage degradation model parameters are identified offline on the basis of an expectation maximization algorithm; the offline obtained parameter estimation values and statistical characteristics of the change-point distribution serve as prior information of online parameter updating; the step of on-line parameter estimation and remaining life prediction includes the substeps that degradation data is collected online; online updating based on Bayesian theory model parameters is conducted; the remaining life of currently running equipment is estimated based on the updated parameters. According to the method, the degradation data with two-stage characteristics can be modeled, and the remaining life of the degradation data can be accurately predicted.

The invention discloses a Web page access forecasting method based on a k-order hybrid Markov model, and the method comprises the following steps of: firstly, gathering and handling Web server access log data, identifying a client and users and eliminating insignificant access data; then identifying user sessions to construct a Web log database; select log data from the database according to forecasting targets, and organizing (k+1) tuples by taking a session as a unit, wherein the (k+1) tuples are used for training the k-order hybrid Markov model; learning and calibrating a parameter set of the k-order hybrid Markov model by using an expectation-maximization algorithm; and identifying sessions according to page access operations of target users and applying the model for forecasting Web pages to be accessed by the users at the next step. The method provided by the invention can be used for recommending pages needed to be accessed by the users so as to reduce the delay of page access and optimize user experience; and in addition, from the point of a Web server, the organizational structure of the Web pages can be improved, the result sequence of a search engine is guided, and the page cache mechanism can be improved, thereby enhancing the quality of service.

A method and system for distributed tracking of multiple targets is disclosed. Multiple targets to be tracked by a plurality of trackers are detected in a frame. The motion state variable of each of the plurality of trackers is calculated in the E-step of a variational Expectation-Maximization algorithm. Further, the data association variable of each of the plurality of trackers is calculated in the M-step of the algorithm. Depending on the motion state variable and the data association variable, the multiple targets are tracked.

A probability distribution for speech model parameters, such as auto-regression parameters, is used to identify a distribution of denoised values from a noisy signal. Under one embodiment, the probability distributions of the speech model parameters and the denoised values are adjusted to improve a variational inference so that the variational inference better approximates the joint probability of the speech model parameters and the denoised values given a noisy signal. In some embodiments, this improvement is performed during an expectation step in an expectation-maximization algorithm. The statistical model can also be used to identify an average spectrum for the clean signal and this average spectrum may be provided to a speech recognizer instead of the estimate of the clean signal.

The invention discloses a music similarity detection method based on mixed characteristics and a Gaussian mixed model. According to the basic thought, the method comprises the steps of using a gamma-tone cepstrum coefficient for conducting similarity detection, and using weighting similarities of various characteristics as a final detection result; providing a modulation spectrum characteristic based on a frame shaft, using the characteristic for representing a music long-time characteristic, and using the combination of the long-time characteristic and a short-time characteristic as the input of modeling in the next step; using the Gaussian mixed model for conducting modeling on the music characteristics, firstly, utilizing a dynamic K mean value method for conducting initialization on the model, then, using an expectation-maximization algorithm for conducting model training, obtaining accurate model parameters, and finally using a log-likelihood ratio algorithm for obtaining the similarities between the pieces of music. According to the music similarity detection method, the obtaining of the music characteristics is more sufficient and thorough, the accuracy degree of music recommendation is improved, the characteristic vector dimensionality can be reduced, the information memory content of a music database is reduced, and the accuracy degree of the music recommendation is improved.

The invention provides a random time delay modeling method based on a continuous time hidden Markov model for a network control system. A network control system is established according to the Figure 1, wherein a controlled object is a 4-degree-of-freedom mechanical arm, and kinematics parameters of the 4-degree-of-freedom mechanical arm are sampled and then transmitted to a controller via the network; the controller firstly establishes a continuous hidden Markov model of network time delay, then, designs an appropriate control law according to the sampled data and transmits the control law to an executor so as to compensate the influence of time delay on system performance. An incomplete data expectation maximization algorithm is used for model parameter training, and simultaneously, a genetic algorithm and a simulated annealing algorithm are both used for ensuring that the training process can quickly converge to a global extremum. Finally, the established network time delay continuous hidden Markov model is verified and optimized on a simulation experiment platform. By the adoption of the method, the time delay generation mechanization that the network time delay is controlled by the network state can be accurately described, and the maximum degree of approximation of the time delay model to the actual network environment can be guaranteed.

The invention provides a real-time wind power prediction and error analysis method based on mixture Gaussian distribution.The real-time wind power prediction and error analysis method is characterized by including the steps of data acquisition and processing, mixed Gaussian distribution model establishment, calculation of mixed distribution function parameters and model evaluation based on an expectation maximization algorithm and the like, linear combination is conducted on multiple Gaussian distributions to obtain a 'multi-weight mixed Gaussian distribution model', model parameters are estimated by adopting the expectation maximization algorithm, and the computational complexity of the maximum likelihood estimation is reduced; modeling is performed by directly relying on the objective law of error distributions themselves, man-made subjectivity is avoided, the shortcoming that a single distribution model is inflexible in shape and poor in universality is overcome, and more accurate description to real-time wind power prediction errors is achieved.The real-time wind power prediction and error analysis method has the advantage of being scientific, reasonable, good in applicability, high in prediction accuracy and the like.

The invention discloses a remote sensing image alteration detection method comprising the steps that two time phase high-resolution optical remote sensing images X1 and X2 are obtained; image registration is performed on X1 and X2; radiation normalization correction is performed on X1 and X2 by utilizing a multivariate alteration detection method; alteration vector amplitude XM and spectral angle information XSA are respectively calculated according to X1 and X2 after radiation normalization correction; an optimal segmentation threshold T is calculated by utilizing the Bayes theory and an expectation maximization algorithm according to the XM; a pseudo training sample area is selected according to T and XM; XM and XSA are combined to act as input of a core FCM, and optimal model parameter value selection is performed on a core FCM combining spatial neighborhood information model according to the pseudo training sample area; and the alteration area and the non-alteration area of the optical remote sensing images are determined by adopting the method of core FCM combining spatial neighborhood information according to the selected optimal model parameter values. The remote sensing image alteration detection method is more robust and higher in precision.

The invention relates to an industrial process fault detection method based on an autoregression dynamic hidden variable model. According to the method, by use of data at a normal operation state in an industrial process, a dynamical process model with universality is established, and the model predicts a model structure through a Kalman filter and a smoother and an expectation maximization algorithm, such that dynamical and static coupling relation of industrial data can be effectively extracted; and then, based on dynamic and static noise of the dynamic model, a corresponding monitoring statistical amount is constructed, and a final fault detection result is obtained. Compared with other existing methods, the method provided by the invention can greatly improve the effects of industrial process dynamic modeling and fault detection, reduces the false alarm rate and the missed alarm rate of faults, improves the monitoring performance to a quite large degree, enhances the understanding capability and the operation confidence of process operators for the process and better facilitates automation implementation of the industrial process.

The invention relates to a moving object detection method based on improved mixing gauss and image cutting, which belongs to the technical field of intelligent video monitoring, and aims to solve the problems of the existing method and improve the accuracy and the processing speed of moving object detection. The method comprises the following steps of: (1) confirming a moving area, wherein the invention provides a three-frame difference and curve fitting method based on gray level difference so as to preliminarily confirm the moving area of an object; (2) initializing an expectation-maximization (EM) algorithm, and confirming an initial value of the EM algorithm by utilizing a method based on grid and density estimation so as to reduce the dependency of the EM algorithm on the initial value; and (3) detecting the moving object, utilizing the initialized EM algorithm to estimate a mixing gauss model parameter, and only adopting a mixing gauss model method to detect the moving object on a cut moving area, so that the computation complexity is greatly reduced. According to the moving object detection method based on improved mixing gauss and image cutting, the accuracy of object detection can be ensured, and simultaneously, the algorithm can meet the performance requirement of actual application.

The invention provides a co-primer array non-grid DOA estimation method under a non-negative sparse Bayes learning framework, and belongs to the field of research on a high-resolution direction finding method in signal processing. The method includes the steps that firstly, a co-primer array received data covariance matrix is vectorized, a virtual received signal model is established, then a non-negative sparse Bayes model is established based on the characteristic that virtual incident signal elements in the model are not negative, hyper-parameters and a grid point set are iteratively updatedthrough an expectation-maximization algorithm, finally, a signal power spectrum is established according to the finally-updated grid point set and the finally-updated hyper-parameters, and then an estimated DOA is determined through spectrum peak searching. By means of the method, the operation process is converted to a real number field from a complex number field, and therefore the computationcomplexity can be reduced to a certain degree. In addition, by the application of a co-primer array, undetermined DOA estimation can be achieved, the limitation of the number of array elements in themaximum estimable information source number is broken through, thus, the hardware cost can be reduced to a certain degree, and certain engineering application value is achieved.

The present invention relates to an underwater acoustic OFDM time-varying channel estimation method based on sparse Bayesian learning. The method comprises the following steps of: the step 1: inputting channel estimation parameters, wherein the parameters comprise a receiving symbol vector YP, a dictionary matrix [Phi]p, the maximum iterations rmax, a termination threshold e and a noise variance [Sigma]2; the step 2: initializing a hyper-parameter matrix [Gamma] and an iteration count r; the step 3: employing the expectation maximization algorithm to perform solution of a hyper parameter [gamma]; the step 4: performing iteration terminal condition determination, wherein if r<rmax and a formula is met as shown in the description, r=r+1, and the step 3 is returned, if r<rmax and a formula ismet as shown in the description, iteration is terminated, and if r>=rmax, the iteration is terminated; and the step 5: outputting an estimation parameter, comprising sparse channel vector estimationand hyper parameter vector estimation. Compared to a current CS method, the method provided by the invention can improve the precision of the channel estimation, can reduce the system error rate, andhas an actual application value in the actual underwater acoustic OFDM communication system.

The invention provides an uplink channel estimation method of a large-scale MIMO system, and the method comprises the following steps: (1) modelling for a probability model of a channel through usinga Gaussian mixture model; (2) performing channel estimation through using optimal Bayesian parameter estimation; (3) giving an iterative initial value through using a hierarchical clustering algorithm; (4) solving an edge probability density function in the step (2) through using an approximate message passing algorithm; (5) iteratively solving parameters of the Gaussian mixture model through using an expectation maximization algorithm. In the method, sparse characteristics of channel gain in a beam domain are used fully, the Bayesian parameter estimation method is used, learning statistical information in advance is not needed, and compared with the conventional channel estimation based on LS, better MSE performance can be obtained.

A method and computer program product for reducing noise in diffusion tensor imaging data utilizes an optimization function performed on voxels or pixels of the image. The optimization function includes a neighborhood relevance component that considers diffusion tensor matrices from neighboring pixels or voxels. The neighborhood relevance component is modeled as a Markov Random Field. The method includes an iteration that seeks convergence of an Expectation Maximization algorithm performed on the diffusion tensor matrices.

The invention discloses an image set unsupervised co-segmentation method based on deformable graph structure representation. The method is divided into the following four steps: firstly, superpixel segmentation is carried out on each graph in a graph set and a descriptor is extracted; secondly, an inner graph based on a Biharmonic distance is built for each picture, specifically, a two-dimensional image is stretched as a three-dimensional grid, and a Laplacian matrix is established on the three-dimensional grid according to a Riemann flow pattern space manner; thirdly, a deformable hypergraph which includes all superpixel in the graph set is established after the inner graph is established and the Biharmonic distance is calculated in a descriptor space(characteristic space); fourthly, clustering ideas considering connectivity are used, and an energy function which comprises the inner graph, the hypergraph and segmentation size constraints is established. Optimization is carried out by using the expectation maximization algorithm to obtain a final co-segmentation result. The algorithm partly uses CPU(Central Processor Unit) parallelism with CPU parallelism, the co-segmentation method has excellent segmentation accuracy and efficiency performance for the large-scale graph set.